cuFasterTucker: A Stochastic Optimization Strategy for Parallel Sparse FastTucker Decomposition on GPU Platform

• DOI: 10.1145/3648094
• Publication Date: 2024-02-16T12:38:14Z
• Authors: Zixuan Li; Yunchuan Qin; Qi Xiao; Wangdong Yang; Kenli Li
• Keywords: Tucker Decomposition; Rank (graph theory); Speedup

The amount of scientific data is currently growing at an unprecedented pace, with tensors being a common form that display high-order, high-dimensional, and sparse features. While tensor-based analysis methods are effective, the vast increase in size has made processing original tensor infeasible. Tensor decomposition offers solution by decomposing into multiple low-rank matrices or can be efficiently utilized methods. One such algorithm Tucker decomposition, which decomposes N -order factor core tensor. However, many techniques generate large intermediate variables require significant computational resources, rendering them inadequate for high-order high-dimensional tensors. This article introduces FasterTucker novel approach to builds on FastTucker variant decomposition. We propose efficient parallel algorithm, called cuFasterTucker, designed run GPU platform. Our low storage requirements provides effective Compared state-of-the-art algorithms, our achieves speedup approximately 7 23 times.